1. Patchew
  2. linux
  3. View series

[PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.

Sebastian Andrzej Siewior posted 1 patch 1 year, 10 months ago
There is a newer version of this series
include/linux/rwbase_rt.h |  4 ++--
include/linux/rwsem.h     | 11 ++++++++---
2 files changed, 10 insertions(+), 5 deletions(-)
[PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Sebastian Andrzej Siewior 1 year, 10 months ago 
The commit cited below broke the build for PREEMPT_RT because
rwsem_assert_held_write_nolockdep() passes a struct rw_semaphore but
rw_base_assert_held_write() expects struct rwbase_rt. Fixing the type
alone leads to the problem that WARN_ON() is not found because bug.h is
missing.

In order to resolve this:
- Keep the assert (WARN_ON()) in rwsem.h (not rwbase_rt.h)
- Add the rwsem_held_write() which returns true if the lock is already
  write-locked and no reader are around.
- Make rw_base_is_write_locked() do the implementation specific
  (rw_base) writer check.
- Replace the "inline" with __always_inline which was used before.

Fixes: f70405afc99b1 ("locking: Add rwsem_assert_held() and rwsem_assert_held_write()")
Reported-by: Clark Williams <williams@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/rwbase_rt.h |  4 ++--
 include/linux/rwsem.h     | 11 ++++++++---
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/include/linux/rwbase_rt.h b/include/linux/rwbase_rt.h
index 29c4e4f243e47..f2394a409c9d5 100644
--- a/include/linux/rwbase_rt.h
+++ b/include/linux/rwbase_rt.h
@@ -31,9 +31,9 @@ static __always_inline bool rw_base_is_locked(const struct rwbase_rt *rwb)
 	return atomic_read(&rwb->readers) != READER_BIAS;
 }
 
-static inline void rw_base_assert_held_write(const struct rwbase_rt *rwb)
+static __always_inline bool rw_base_is_write_locked(const struct rwbase_rt *rwb)
 {
-	WARN_ON(atomic_read(&rwb->readers) != WRITER_BIAS);
+	return atomic_read(&rwb->readers) == WRITER_BIAS;
 }
 
 static __always_inline bool rw_base_is_contended(const struct rwbase_rt *rwb)
diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
index 4f1c18992f768..c81630d81018d 100644
--- a/include/linux/rwsem.h
+++ b/include/linux/rwsem.h
@@ -167,14 +167,19 @@ static __always_inline int rwsem_is_locked(const struct rw_semaphore *sem)
 	return rw_base_is_locked(&sem->rwbase);
 }
 
-static inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
+static __always_inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
 {
 	WARN_ON(!rwsem_is_locked(sem));
 }
 
-static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
+static __always_inline bool rwsem_held_write(const struct rw_semaphore *sem)
 {
-	rw_base_assert_held_write(sem);
+	return rw_base_is_write_locked(&sem->rwbase);
+}
+
+static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
+{
+	WARN_ON(!rwsem_held_write(sem));
 }
 
 static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)
-- 
2.43.0
Re: [PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Matthew Wilcox 1 year, 10 months ago 
On Tue, Mar 19, 2024 at 08:05:50AM +0100, Sebastian Andrzej Siewior wrote:
> -static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> +static __always_inline bool rwsem_held_write(const struct rw_semaphore *sem)

The locking maintainers were very clear that this predicate Should Not
Exist.  It encourages people to write bad code.  Assertions only!

>  {
> -	rw_base_assert_held_write(sem);
> +	return rw_base_is_write_locked(&sem->rwbase);
> +}
> +
> +static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> +{
> +	WARN_ON(!rwsem_held_write(sem));
>  }
>  
>  static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)
> -- 
> 2.43.0
>
Re: [PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Sebastian Andrzej Siewior 1 year, 10 months ago 
On 2024-03-19 13:38:06 [+0000], Matthew Wilcox wrote:
> On Tue, Mar 19, 2024 at 08:05:50AM +0100, Sebastian Andrzej Siewior wrote:
> > -static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> > +static __always_inline bool rwsem_held_write(const struct rw_semaphore *sem)
> 
> The locking maintainers were very clear that this predicate Should Not
> Exist.  It encourages people to write bad code.  Assertions only!

What do you refer to? The inline vs __always_inline or
rwsem_held_write() should not exists and it should invoke directly
rw_base_is_write_locked()?

> >  {
> > -	rw_base_assert_held_write(sem);
> > +	return rw_base_is_write_locked(&sem->rwbase);
> > +}
> > +
> > +static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> > +{
> > +	WARN_ON(!rwsem_held_write(sem));
> >  }
> >  
> >  static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)

Sebastian
Re: [PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Matthew Wilcox 1 year, 10 months ago 
On Tue, Mar 19, 2024 at 03:15:06PM +0100, Sebastian Andrzej Siewior wrote:
> On 2024-03-19 13:38:06 [+0000], Matthew Wilcox wrote:
> > On Tue, Mar 19, 2024 at 08:05:50AM +0100, Sebastian Andrzej Siewior wrote:
> > > -static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> > > +static __always_inline bool rwsem_held_write(const struct rw_semaphore *sem)
> > 
> > The locking maintainers were very clear that this predicate Should Not
> > Exist.  It encourages people to write bad code.  Assertions only!
> 
> What do you refer to? The inline vs __always_inline or
> rwsem_held_write() should not exists and it should invoke directly
> rw_base_is_write_locked()?

I see Waiman already gave you the substantial answer ... but why did you
change inline to __always_inline?
Re: [PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Sebastian Andrzej Siewior 1 year, 10 months ago 
On 2024-03-19 16:01:26 [+0000], Matthew Wilcox wrote:
> I see Waiman already gave you the substantial answer ... but why did you
> change inline to __always_inline?

To align with the all the other functions in the file which were defined
like that.

Sebastian
Re: [PATCH] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Waiman Long 1 year, 10 months ago 
On 3/19/24 10:15, Sebastian Andrzej Siewior wrote:
> On 2024-03-19 13:38:06 [+0000], Matthew Wilcox wrote:
>> On Tue, Mar 19, 2024 at 08:05:50AM +0100, Sebastian Andrzej Siewior wrote:
>>> -static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
>>> +static __always_inline bool rwsem_held_write(const struct rw_semaphore *sem)
>> The locking maintainers were very clear that this predicate Should Not
>> Exist.  It encourages people to write bad code.  Assertions only!
> What do you refer to? The inline vs __always_inline or
> rwsem_held_write() should not exists and it should invoke directly
> rw_base_is_write_locked()?

Just merge rwsem_held_write() into rwsem_assert_held_write_nolockdep() 
and we should be all set.

Cheers,
Longman

>>>   {
>>> -	rw_base_assert_held_write(sem);
>>> +	return rw_base_is_write_locked(&sem->rwbase);
>>> +}
>>> +
>>> +static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
>>> +{
>>> +	WARN_ON(!rwsem_held_write(sem));
>>>   }
>>>   
>>>   static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)
> Sebastian
>
[PATCH v2] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Sebastian Andrzej Siewior 1 year, 10 months ago 
The commit cited below broke the build for PREEMPT_RT because
rwsem_assert_held_write_nolockdep() passes a struct rw_semaphore but
rw_base_assert_held_write() expects struct rwbase_rt. Fixing the type
alone leads to the problem that WARN_ON() is not found because bug.h is
missing.

In order to resolve this:
- Keep the assert (WARN_ON()) in rwsem.h (not rwbase_rt.h)
- Make rwsem_assert_held_write_nolockdep() do the implementation
  specific (rw_base) writer check.
- Replace the "inline" with __always_inline which was used before.

Fixes: f70405afc99b1 ("locking: Add rwsem_assert_held() and rwsem_assert_held_write()")
Reported-by: Clark Williams <williams@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
v1…v2:
  - Merge rwsem_held_write() into rwsem_assert_held_write_nolockdep() as
    per Waiman Long / Matthew Wilcox.

 include/linux/rwbase_rt.h | 4 ++--
 include/linux/rwsem.h     | 6 +++---
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/include/linux/rwbase_rt.h b/include/linux/rwbase_rt.h
index 29c4e4f243e47..f2394a409c9d5 100644
--- a/include/linux/rwbase_rt.h
+++ b/include/linux/rwbase_rt.h
@@ -31,9 +31,9 @@ static __always_inline bool rw_base_is_locked(const struct rwbase_rt *rwb)
 	return atomic_read(&rwb->readers) != READER_BIAS;
 }
 
-static inline void rw_base_assert_held_write(const struct rwbase_rt *rwb)
+static __always_inline bool rw_base_is_write_locked(const struct rwbase_rt *rwb)
 {
-	WARN_ON(atomic_read(&rwb->readers) != WRITER_BIAS);
+	return atomic_read(&rwb->readers) == WRITER_BIAS;
 }
 
 static __always_inline bool rw_base_is_contended(const struct rwbase_rt *rwb)
diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
index 4f1c18992f768..c8b543d428b0a 100644
--- a/include/linux/rwsem.h
+++ b/include/linux/rwsem.h
@@ -167,14 +167,14 @@ static __always_inline int rwsem_is_locked(const struct rw_semaphore *sem)
 	return rw_base_is_locked(&sem->rwbase);
 }
 
-static inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
+static __always_inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
 {
 	WARN_ON(!rwsem_is_locked(sem));
 }
 
-static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
+static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
 {
-	rw_base_assert_held_write(sem);
+	WARN_ON(!rw_base_is_write_locked(&sem->rwbase));
 }
 
 static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)
-- 
2.43.0
Re: [PATCH v2] Locking: Let PREEMPT_RT compile again with new rwsem asserts.
Posted by Waiman Long 1 year, 10 months ago 
On 3/19/24 14:20, Sebastian Andrzej Siewior wrote:
> The commit cited below broke the build for PREEMPT_RT because
> rwsem_assert_held_write_nolockdep() passes a struct rw_semaphore but
> rw_base_assert_held_write() expects struct rwbase_rt. Fixing the type
> alone leads to the problem that WARN_ON() is not found because bug.h is
> missing.
>
> In order to resolve this:
> - Keep the assert (WARN_ON()) in rwsem.h (not rwbase_rt.h)
> - Make rwsem_assert_held_write_nolockdep() do the implementation
>    specific (rw_base) writer check.
> - Replace the "inline" with __always_inline which was used before.
>
> Fixes: f70405afc99b1 ("locking: Add rwsem_assert_held() and rwsem_assert_held_write()")
> Reported-by: Clark Williams <williams@redhat.com>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> ---
> v1…v2:
>    - Merge rwsem_held_write() into rwsem_assert_held_write_nolockdep() as
>      per Waiman Long / Matthew Wilcox.
>
>   include/linux/rwbase_rt.h | 4 ++--
>   include/linux/rwsem.h     | 6 +++---
>   2 files changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/rwbase_rt.h b/include/linux/rwbase_rt.h
> index 29c4e4f243e47..f2394a409c9d5 100644
> --- a/include/linux/rwbase_rt.h
> +++ b/include/linux/rwbase_rt.h
> @@ -31,9 +31,9 @@ static __always_inline bool rw_base_is_locked(const struct rwbase_rt *rwb)
>   	return atomic_read(&rwb->readers) != READER_BIAS;
>   }
>   
> -static inline void rw_base_assert_held_write(const struct rwbase_rt *rwb)
> +static __always_inline bool rw_base_is_write_locked(const struct rwbase_rt *rwb)
>   {
> -	WARN_ON(atomic_read(&rwb->readers) != WRITER_BIAS);
> +	return atomic_read(&rwb->readers) == WRITER_BIAS;
>   }
>   
>   static __always_inline bool rw_base_is_contended(const struct rwbase_rt *rwb)
> diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
> index 4f1c18992f768..c8b543d428b0a 100644
> --- a/include/linux/rwsem.h
> +++ b/include/linux/rwsem.h
> @@ -167,14 +167,14 @@ static __always_inline int rwsem_is_locked(const struct rw_semaphore *sem)
>   	return rw_base_is_locked(&sem->rwbase);
>   }
>   
> -static inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
> +static __always_inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
>   {
>   	WARN_ON(!rwsem_is_locked(sem));
>   }
>   
> -static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
> +static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
>   {
> -	rw_base_assert_held_write(sem);
> +	WARN_ON(!rw_base_is_write_locked(&sem->rwbase));
>   }
>   
>   static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)
Reviewed-by: Waiman Long <longman@redhat.com>
[tip: locking/urgent] locking: Make rwsem_assert_held_write_nolockdep() build with PREEMPT_RT=y
Posted by tip-bot2 for Sebastian Andrzej Siewior 1 year, 10 months ago 
The following commit has been merged into the locking/urgent branch of tip:

Commit-ID:     fa1f51162338b3e2f520d4bfedc42b3b2e00da6d
Gitweb:        https://git.kernel.org/tip/fa1f51162338b3e2f520d4bfedc42b3b2e00da6d
Author:        Sebastian Andrzej Siewior <bigeasy@linutronix.de>
AuthorDate:    Tue, 19 Mar 2024 19:20:50 +01:00
Committer:     Thomas Gleixner <tglx@linutronix.de>
CommitterDate: Mon, 08 Apr 2024 16:39:16 +02:00

locking: Make rwsem_assert_held_write_nolockdep() build with PREEMPT_RT=y

The commit cited below broke the build for PREEMPT_RT because
rwsem_assert_held_write_nolockdep() passes a struct rw_semaphore but
rw_base_assert_held_write() expects struct rwbase_rt. Fixing the type alone
leads to the problem that WARN_ON() is not found because bug.h is missing.

In order to resolve this:

 - Keep the assert (WARN_ON()) in rwsem.h (not rwbase_rt.h)

 - Make rwsem_assert_held_write_nolockdep() do the implementation
   specific (rw_base) writer check.

 - Replace the "inline" with __always_inline which was used before.

Fixes: f70405afc99b1 ("locking: Add rwsem_assert_held() and rwsem_assert_held_write()")
Reported-by: Clark Williams <williams@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Waiman Long <longman@redhat.com>
Link: https://lore.kernel.org/r/20240319182050.U4AzUF3I@linutronix.de
---
 include/linux/rwbase_rt.h | 4 ++--
 include/linux/rwsem.h     | 6 +++---
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/include/linux/rwbase_rt.h b/include/linux/rwbase_rt.h
index 29c4e4f..f2394a4 100644
--- a/include/linux/rwbase_rt.h
+++ b/include/linux/rwbase_rt.h
@@ -31,9 +31,9 @@ static __always_inline bool rw_base_is_locked(const struct rwbase_rt *rwb)
 	return atomic_read(&rwb->readers) != READER_BIAS;
 }
 
-static inline void rw_base_assert_held_write(const struct rwbase_rt *rwb)
+static __always_inline bool rw_base_is_write_locked(const struct rwbase_rt *rwb)
 {
-	WARN_ON(atomic_read(&rwb->readers) != WRITER_BIAS);
+	return atomic_read(&rwb->readers) == WRITER_BIAS;
 }
 
 static __always_inline bool rw_base_is_contended(const struct rwbase_rt *rwb)
diff --git a/include/linux/rwsem.h b/include/linux/rwsem.h
index 4f1c189..c8b543d 100644
--- a/include/linux/rwsem.h
+++ b/include/linux/rwsem.h
@@ -167,14 +167,14 @@ static __always_inline int rwsem_is_locked(const struct rw_semaphore *sem)
 	return rw_base_is_locked(&sem->rwbase);
 }
 
-static inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
+static __always_inline void rwsem_assert_held_nolockdep(const struct rw_semaphore *sem)
 {
 	WARN_ON(!rwsem_is_locked(sem));
 }
 
-static inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
+static __always_inline void rwsem_assert_held_write_nolockdep(const struct rw_semaphore *sem)
 {
-	rw_base_assert_held_write(sem);
+	WARN_ON(!rw_base_is_write_locked(&sem->rwbase));
 }
 
 static __always_inline int rwsem_is_contended(struct rw_semaphore *sem)

Patchew 2.1-devel-b67e4c2

© 2016 - 2026 Red Hat, Inc.